On Security And Temperature Aware Dynamic Resource Management Methods For Many-Core Systems

With the coming of the post-Moore era,increasing the integration scale of a single chip(2D integrated packaging)reduces chip yields as well as increases chip manufacturing costs,so using advanced packaging technologies(3D integrated packaging and 2.5D multi-chiplet integrated packaging)improves chip computing power.The integration density of many-core systems is increasing,and the power density and thermal correlation of the chips are increasing.Therefore,the security and performance of many-core systems will face the following challenges: First,attackers can transmit information through chip temperature variations,thus leak sensitive data,which is referred as Thermal Covert Channel(TCC)attack.TCC is stealthy and poses a serious threat to the information security of many-core systems.Second,because of the high power density of many-core systems,hot spots(high temperature in a certain area)are easily generated when the application is running in the system,which affects the system reliability.To ensure the reliability of the system operation,system performance are optimized within safety and temperature constraints.Since dynamic resource management methods(task migration and task mapping)determine the thermal distribution and communication overhead of a many-core system,and thus are key to optimize the system performance.For 2D many-core systems and 3D integrated systems,this dissertation models the TCC BER and the thermal power budget of cores,and proposes a task migration based dynamic resource management method with security and temperature constraints to optimize system performance.Simulation experiments are conducted on 2D many-core system and 3D integrated system,which have different architectural characteristics.The experimental results show that the task migration based method increases the average packet error rate of TCC to84% and effectively blocks TCC data transmission.Current defense methods against TCC attacks include dynamic voltage frequency scaling and noise jamming,but these methods cause system performance degradation or increase in power consumption.Experimental results show that the proposed method has the lowest peak temperature and reduces system performance loss by 56% compared to existing methods.In addition,the communication characteristics of 2.5D multi-chiplet based many-core systems are different from 2D many-core systems and 3D integrated systems: the inter-chiplet communication delay is much larger than the intra-chiplet communication delay.Therefore,this dissertation proposes a dynamic resource management method to perform task mapping for 2.5D multi-chiplet based many-core systems.The goal is minimizing system communication overhead and the constraints are security and temperature.Existing task mapping algorithms applied to 2.5D multi-chiplet based many-core systems do not consider their special communication characteristics and thus cause performance loss.Simulation experiments are conducted on 2.5D multi-chiplet based many-core systems.Experimental results show that the proposed method increases the average packet error rate of TCC to 85.6%and effectively blocks TCC data transmission in 2.5D multi-chiplet based many-core systems.Compared to existing methods,the proposed method has the lowest peak temperature and reduced the application runtime by 48.2% and the communication latency by 38.6%.